How Active learning works?

  1. Does active learning help me to get the next annotation sample based on my previous annotation? How frequently gradients get updated in active learning?

  2. For the NER case, Does my prediction for entities get smarter with my immediate previous annotation with active learning?

3, Do you provide any kind of free version for trial? We are really interested to try this out.

Incase already answered somewhere, please provide the link.

Annotations are sent back in batches of batch_size (10 by default). You can customise this in the settings and also implement your own updating mechanism that updates

This is typically not what you want – you want to update your model in small batches and give it time to update in the background as you annotate more data. Because streams are Python generators and processed in batches, updates to the model are reflected in future batches it processes.

It's not that helpful to think of it as "smarter", though :wink: The model will be updated and its weights will change and hopefully produce more accurate predictions.

We normally do commercial trials by hosting a VM, which lets people test the full tool and the scriptable back-end. You can email us at Also see the FAQs for details:

Thank you.